Machine Learning and Knowledge Based Methods to Model Wildfire Susceptibility in Belize’s Ecosystems and Protected Areas
Santos CHICAS, IRI-THESys and Geography Department, Humboldt-Universität, Germany
Miguel C. VALDEZ, Center for Space and Remote Sensing Research, National Central University, Taiwan
Jonas Ø. NIELSEN, IRI-THESys and Geography Department, Humboldt-Universität, Germany
Wildfires are serious environmental threats around the world, especially in the Global South where resources to suppress and mitigate the impacts of wildfires are limited. As a result, it is a necessity for managers to have a clear understanding of the spatial distribution of wildfire vulnerable areas in order to improve fire prevention strategies. Researchers have endeavored to aid in identifying susceptible areas by utilizing models; thus, this study evaluates the ability of Analytical Hierarchical Process (AHP), Fuzzy Analytical Hierarchical Process (FAHP) and Random Forest (RF) to predict wildfire susceptibility in Belize. This generated important model performance information that can be used by other researchers when deciding on wildfire susceptibility model selection. The results of this research indicate that RF is the model with better predictive accuracy with an AUC value of 83.1 followed by FAHP (AUC=71.2) and AHP (AUC= 66.8). The main factors that are influencing wildfire susceptibility in Belize are distance to agriculture, landcover and temperature. The most vulnerable areas are unprotected areas, the outer boundaries of protected areas and small and isolated protected areas with lowland broad-leaved moist forest and lowland savanna being the most vulnerable ecosystems. An inclusion of fire experts based in Belize in identifying the main factors of wildfire susceptibility, ranking of factors and sharing their experience was crucial for this research. This research not only provides valuable information to Belize’s wildfire managers but also contributed by providing an insight of the implementation and performance of these modelling methods in a country were local data availability, accessibility and reliability are an issue and where wildfire risk data is lacking.
Mots clés : Wildfires |Risk|Knowledge-Based Method|Protected Area|Belize
A104752SC